Move the ROCm devcontainer to ROCm 7.14 / py3.14 / torch 2.12 with AITER 0.1.16 - #294

Open

demandal25 wants to merge 50 commits into

amd-integrationfrom

devcontainer-rocm714

Commits on Aug 27, 2026

feat(rocm): support ROCm 7.14 / py3.14 / torch 2.12 with AITER 0.1.16

demandal25
and
claude
committed
0097252
fix(aiter): propagate the torch stream into AITER's POD entry point

demandal25
and
claude
committed
193b115
fix(rocm): restore build backend deps and harden the 0.1.16 shims
demandal25
committed
ee0a0ce
fix(aiter): route causal soft-cap away from AITER's miscomputing kernel
demandal25
committed
516b866
docs(rocm): describe the ROCm 7.14 / py3.14 / torch 2.12 stack
demandal25
committed
730a7e8
docs(rocm): pass the host's numeric render GID, not the group name
demandal25
committed
9b7fe67
fix(aiter): extend the soft-cap fallback to the batch prefill wrappers
demandal25
committed
4ff8bcc
fix(rocm): drop the scikit-build-core deps the setuptools switch made dead
demandal25
committed
89f9835
fix(rocm): remove the default ubuntu user before creating devuser
demandal25
committed
5722844
fix(rocm): graft ROCm headers and dev symlinks into the devcontainer
demandal25
committed
9e2d1dd
test(rocm): assert single-prefill routing against the single-prefill row
demandal25
committed
495378e
fix(aiter): scope the AITER stream override to the call
demandal25
committed
348e019
fix(aiter): set GPU_ARCHS for AITER's own JIT, not just our shim build
demandal25
committed
02dd016
fix(aiter): port the paged-append shim to AITER's POD API
demandal25
committed
542be2d
fix(aiter): track moe_ck.h's new is_shuffled parameter
demandal25
committed
bc279ad
fix(aiter): close the review gaps in the soft-cap guard and the graft
demandal25
committed
1de0243
fix(aiter): key the paged soft-cap guard on the paging route
demandal25
committed
6b866bf
fix(aiter): let the shim build's GPU_ARCHS outlive its own scope
demandal25
committed
534b360
docs(rocm): stop claiming the base image sets the ROCm env vars
demandal25
committed
99a6068
fix(aiter): reject non-NHD kv_layout on explicit single-prefill aiter
demandal25
committed
e8f1a0a
fix(aiter): enforce an amd-aiter 0.1.16 ABI floor before routing
demandal25
committed
63a7b64
docs(skills): drop line-number anchors that no longer resolve
demandal25
committed
cbe8f0d
fix(aiter): port the fp8 per-token quant call to AITER's POD API
demandal25
committed
1e4a5d7
fix(aiter): run the fp8 quant kernel on the caller's stream
demandal25
committed
d4d03a9
Merge branch 'amd-integration' into devcontainer-rocm714
demandal25
authored
f20faa1
Merge branch 'amd-integration' into devcontainer-rocm714
demandal25
authored
ad74fd8
fix(rocm): only check PyTorch's arch list when PYTORCH_ROCM_ARCH pins it
demandal25
committed
f7bae0d
docs(rocm): describe the arch check as the PYTORCH_ROCM_ARCH gate it is
demandal25
committed
84c0e3c
fix(aiter): compare AITER versions without stripping the pre-release
demandal25
committed
2ce803f
fix(rocm): re-guard the soft cap when a native page size degrades
demandal25
committed
f731733
Merge branch 'amd-integration' into devcontainer-rocm714
demandal25
authored
389d662

Commits on Aug 28, 2026